One Board request sounded straightforward enough: explain monthly revenue variance. In the ARM world, we initially approached it using a standard three-factor decomposition:

• Volume (placements)
• Liquidity Rate
• Fee Rate

Simple in theory. Except variance attribution models are rarely as objective as they appear. The problem emerged almost immediately: sequence mattered.

If volume variance is calculated using:

• Actual placement difference × Budget LR × Budget Fee Rate

…you get one answer.

If another faction prefers:

• Actual placement difference × Actual LR × Actual Fee Rate

…you get a different answer.

Both methods are mathematically defensible. Neither is entirely “wrong.” But each embeds a different assumption about causality. That was the real lesson. The discussion quietly shifted from: “What caused the variance?” …to: “Which underlying assumptions are we willing to institutionalize in the dashboard?”

As different Board factions developed preferences for different attribution logic, the model itself became part of the debate. To simplify things, we eventually collapsed LR and Fee Rate into a combined factor and moved from a three-way variance to a two-way model.

That reduced complexity. It did not reduce disagreement. Eventually the compromise became:

• present the KPIs directly,
• show placements AVB,
• show LR AVB,
• show Fee Rate AVB,
• and avoid explicit revenue attribution altogether.

Cleaner politically. Less satisfying analytically. The irony was that simplifying the dashboard reduced the very insight the dashboard was supposed to provide. And naturally, once the next quarter’s results arrived, the attribution debates resurfaced anyway.

The broader lesson: Dashboards are excellent monitoring tools, especially for trend analysis. But once metrics begin assigning causality rather than simply reporting conditions, the underlying assumptions matter enormously. Expose too much detail and organizations can spiral into endless debates over what is “correct.” Expose too little and the true operational drivers become increasingly opaque.

Sometimes the hardest part of dashboard design is not the math. It is deciding which version of reality the organization is prepared to accept.